Video Sequences Of Dynamic Curve Calibration And Brightness Correction

Currently, image-based rendering is a focusing point of research which has attracted a lot of interdisciplinary efforts from both Computer Graphics and Computer Vision communities. Different from traditional Graphics, what can be seen from different viewpoints of a given scenery are generated based on a set of pre-captured photographs. As the technology of image-based rendering progresses, it extends to another domain: image-based lighting, which no longer has the restriction on constant illuminating conditions, so that users can adjust both viewpoint and illumination to reflect the scene in more flexible ways.High Dynamic Range Image (HDRI) is a special kind of image format whose scope for luminance is quite large, with its most important characteristic being that: luminance information stored in an HDRI is linearly correlated to true radiance sensed in real scenes. HDRI is an indispensable constituting part of image-based lighting models, as it realizes the notion of lighting 3-D scenes using valid luminance information. Traditional methods to acquire HDRI make use of pictures taken by a fixed camera under different exposure degrees, so that the camera's radiometric response curve can be calibrated. In this paper, we introduce another algorithm to do it, which uses an automatically exposed video sequence gathered at a fixed observing point. This new strategy does not require the recording of exposure degrees as traditional methods do: it at first calibrates the geometric transformation between frames in the video, then pixel value differences in overlapping regions are calculated for selection of key frames, which are finally input to a routine using polynomial models for response function calibration.As the direct application for recovering response function from a video sequence, we use the response curve calibrated above to solve the problem of pixel value differences in panorama. Panorama is a very useful image-based rendering technique, because of its small data size to represent virtual scenes, and high rendering speed. While people stitch a panorama with a series of photos or a video sequence, the exposure may vary among every frame according to current scene radiance because of auto exposure. This exposure variance often causes different pixel values in different frames of a same point, which leads to discontinuousluminance in the composited panoramic image. We focus on such kind of pixel value differences, effectively integrate HDRI's properties into panorama composition. At first, recovered response function is used to compute the correct radiance of each pixel, and then we can map the radiance of all frames into an image space with the same exposure, correct differences, and achieve color correction.